Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Greptile Summary

This PR adds two documentation-only files: a root AGENTS.md to orient non-Claude coding tools (Codex, GPT, etc.) within the Shakapacker project, and prompts/address-review.md — a shared prompt template that mirrors the existing Claude /address-review slash-command workflow. No Ruby, JavaScript, or test files are touched.

Confidence Score: 5/5

Safe to merge — documentation-only additions with no impact on runtime code.

Both files are purely documentation. The single P2 finding is a placeholder-naming clarity issue in an AI prompt template that doesn't affect any production code path.

prompts/address-review.md — minor placeholder convention inconsistency worth addressing before the template is widely used.

Important Files Changed

Sequence Diagram

sequenceDiagram
    participant Dev as Developer
    participant AI as Coding Assistant (Codex / GPT)
    participant GH as GitHub API (gh)

    Dev->>AI: Address PR review comments for PR N
    AI->>AI: Read prompts/address-review.md
    alt Terminal / gh available
        AI->>GH: gh repo view (get REPO)
        AI->>GH: gh api pulls/PR/comments
        AI->>GH: gh api graphql (thread metadata)
        GH-->>AI: comments + thread state
        AI->>AI: Filter and triage (MUST-FIX / DISCUSS / SKIPPED)
        AI-->>Dev: Present triage list, await selection
        Dev->>AI: Select items to address
        AI->>AI: Apply code changes
        AI->>GH: POST reply comments
        AI->>GH: resolveReviewThread (mutation)
    else No gh access
        AI-->>Dev: Request exported comment data
        Dev->>AI: Provide comment data
        AI->>AI: Triage and address as above
    end

_{Reviews (1): Last reviewed commit: "Sync address-review prompt with upstream..." | Re-trigger Greptile}

Code Review: PR #1098 — Sync address-review prompt with upstream

Overview

Low-risk, docs/tooling-only change. Adds:

1. AGENTS.md — project overview and working rules for non-Claude AI tools (Codex, GPT, etc.)
2. prompts/address-review.md — a standalone prompt template mirroring the Claude /address-review workflow

Both files have proper trailing newlines. No runtime code is touched.

Issues Found

AGENTS.md — missing two rules from CLAUDE.md

The Working Rules section omits two rules that are listed as "Critical Requirements" in CLAUDE.md:

1. Trailing newline rule — CLAUDE.md says "ALWAYS end all files with a trailing newline character. This is required by the project's linting rules." Non-Claude tools creating/editing files won't know this.
2. Dual bundler testing — CLAUDE.md says "Test changes with both bundlers when modifying core functionality" (webpack + rspack). Since the project explicitly supports both, this is worth carrying over.

Suggested additions to the Working Rules section:

- Always end files with a trailing newline (required by project linting rules).
- Test changes with both webpack and rspack bundlers when modifying core functionality.

prompts/address-review.md — Known Limitations not ported

The Claude command (/.claude/commands/address-review.md) includes a "Known Limitations" section covering:

• GitHub API rate limits
• Private repo auth scope requirement
• GraphQL comments(first:100) inner pagination truncation for threads with >100 comments

These caveats are equally relevant to non-Claude tools executing the workflow. Worth adding a short limitations note at the end of the prompt template.

Maintenance Note

AGENTS.md and CLAUDE.md now have overlapping rules. Consider adding a comment in AGENTS.md pointing editors to keep both files in sync, or noting that CLAUDE.md is the canonical source and AGENTS.md is the non-Claude subset.

Summary

The approach is solid and the prompt template faithfully captures the triage-then-act workflow. The two omissions above are minor but worth fixing before merge since they affect correctness for non-Claude users of this repo.

Review

Overview: Adds prompts/address-review.md — a copy-paste prompt template that lets non-Claude assistants (Codex, ChatGPT, etc.) run the same address-review triage workflow. Good idea for cross-tool parity.

Issues

1. PR description is stale after the second commit

The description still says:

Add root AGENTS.md (adapted for Shakapacker's layout)…

And the test plan still has:

• Verify AGENTS.md is discoverable by Codex CLI…

AGENTS.md was dropped in the "Drop AGENTS.md from PR" commit, so these are dead. The description and test plan should be updated to reflect what's actually in the diff.

2. Missing explanation for review-summary-body reply constraint (Step 7)

The prompt correctly lists gh api repos/${REPO}/issues/{PR_NUMBER}/comments for review summary bodies, but gives no reason why. The Claude command spells this out:

"Review summary bodies do not have a comment_id and cannot be replied to via the /replies endpoint."

Without this note, an AI assistant following the prompt could still try the /replies endpoint on a review body (which will 404), since the rule is non-obvious and the why is absent. Suggest adding a one-sentence explanation inline.

3. Multi-line commands in Step 3 are written as inline backtick spans

Several gh api calls in Step 3 are multi-line but formatted as inline backtick code rather than fenced blocks. This makes it easy for a reader (human or AI) to miss the line boundaries and run them incorrectly. The Claude command uses fenced code blocks for every command — this prompt should do the same.

4. No example output

The Claude command includes a concrete example triage block showing the expected format, which helps non-Claude tools produce consistent output. The new prompt omits this; adding even a compact version would improve reliability across different AI backends.

Minor

• No Known Limitations section (rate limits, comments(first:100) inner pagination cap) — low priority but worth carrying forward from the Claude command.
• File ends with a trailing newline ✓

Overall this is low-risk (docs only), but issues 1–3 above are worth fixing before merge.